Full-text search across 69454 articles. Combine with topic and date filters; results sorted by relevance.
Anthropic's launch post for Claude Fable 5.2 and Claude Mythos 5.2 states that during training the company limited advances in research capabilities that could most directly accelerate AI development, so both models perf…
A developer released Vork, a headless 3D and VFX engine built for AI agents, in which agents write entire scenes as code — including their own shaders, materials, cameras and physics — for GPU rendering. The Vork core AP…
A developer built GAWE, an open-source agent engine that enforces a "model proposes, deterministic code decides" rule, and ran 125 prompt-injection attacks against it using Claude Sonnet 5.5 across mock email, CRM, e-sig…
Anthropic announced on November 9 that Claude Fable 5.2 and Claude Mythos 5.2 are available now, priced at $8 per million input tokens — below the predecessor models — and running up to 30% faster than version 5.1. Anthr…
PostEverywhere launched a social media scheduling tool that connects to ChatGPT, Claude, Gemini, Grok, Perplexity, Cursor, Windsurf, Cline, n8n, Claude Code, Codex and Microsoft Copilot Studio through a hosted MCP server…
Meta released Muse Ember on November 9, pricing its smallest Muse model at $0.10 per million input tokens versus $1.25 for Spark, and made it available on the Meta Model API and in Muse Code. Meta CEO Mark Zuckerberg ann…
Handshake's AI training business reached $1 billion in gross annualized revenue roughly a year after launch, Garrett Lord said on a Latent Space deep dive covering what frontier labs buy from expert-data providers. The e…
Since about November 5, Outlook.com, Hotmail and Live addresses have been rejecting some mail from ordinary, low-volume senders with 550 5.7.23 bounces that blame SPF or a missing DMARC record, and on November 9 a Micros…
A blogger rebuilt their personal blog in Rust after AI generated the new site in under an hour from plain-English commands, replacing the Jekyll static site generator they had used for years. The Rust blog runs on a serv…
An experiment planting four classic data-leakage patterns into real datasets found that leakage inflated validation accuracy to 97.0% (AUC 0.999) on an Amazon stock-direction task versus 51.6% (AUC 0.476) with only prior…
UptimeRobot added third party service monitoring to its API, MCP server and CLI, letting teams track outages at more than 6,000 cloud and SaaS providers from code, AI assistants or the terminal. The MCP server gains 10 n…
A developer describes adopting GPTScript, an open-source scripting framework for building AI agents that treats tool calls as first-class language constructs and lets agents be written in shell-script-like syntax rather …
FOSSIL v3.0.0, a pure Fortran geometry library for simulation codes, has been released with surfaces, signed distances and level sets, immersed-boundary queries, AMR tagging, exact cut cells, curvilinear and overset grid…
Apple is testing a macOS 27.3 feature flag named TCC/FDA_WARNING that would stop apps automatically appearing on the Full Disk Access permissions list after a denied read, requiring users to add applications manually, ac…
Simon Willison reported that Claude Opus 5.5 Medium, running in regular Claude Chat, ported a Three.js Nudibranch-themed Fabergé Egg he built into Vanilla JavaScript and WebGL without Three.js, producing a 98KB file vers…
A developer attempting to test whether an agent reasons differently when given today's date found the experiment impossible to run inside Claude Code's agent harness, which automatically injects a "Today's date is 2026-1…
StepFun-affiliated researchers published P²-MOPD, a multi-teacher on-policy distillation method that treats each student module as learnable weights plus a weighted sum of frozen expert task vectors, reporting 94.7% aver…
Production LLM applications should stack a SHA-256 exact-hash cache keyed on normalized prompt, model and tool schema version with an embedding nearest-neighbor tier that returns a prior answer only when similarity clear…
Peking University's DAGroup published Open-HumanWAM, an open recipe for pretraining world–action models on 50,000 hours of egocentric human video and post-training them into robot policies, with the arXiv paper (2611.068…
Meta released SAM 3, an open-weight vision model that performs Promptable Concept Segmentation — returning masks and persistent IDs for every instance matching a short text phrase, exemplar box, or click prompt in both i…